Blender Python Scripting for Visual Effects
Published:
Blender ships with a full Python API (bpy) that exposes nearly every feature of the interface: object creation, material nodes, particle systems, geometry nodes, and rendering.
Posts by Tags
Physics
3D
Beamforming
MEG Beamforming: Forward and Inverse Problems
Published:
Magnetoencephalography (MEG) measures the tiny magnetic fields produced by synchronised postsynaptic currents in cortical neurons. Compared to EEG, MEG is less distorted by the skull and scalp, giving it better spatial resolution.
Blender
Blender Python Scripting for Visual Effects
Published:
Blender ships with a full Python API (bpy) that exposes nearly every feature of the interface: object creation, material nodes, particle systems, geometry nodes, and rendering.
Computational Modeling
Modeling Why the Necker Cube Won’t Sit Still
Published:
Most write-ups of visual illusions stop at showing you the picture: here’s a cube that flips, isn’t the brain weird. That’s true but incomplete, it doesn’t say why perception should flip at all, rather than just settling on one interpretation and staying there, or why the flips arrive when they do rather than on some fixed schedule. This post builds a small computational model of exactly that switching process, for the Necker cube specifically, and asks whether a handful of very simple neural dynamics can reproduce the statistical signature of real perceptual switching.
Computational Neuroscience
My experience at ICLR 2025
Published:
Welcome to my personal recap of attending ICLR 2025 in Singapore! This write-up blends my reflections, highlights from the conference, and a bit of travelogue from one of the most exciting tech and research hubs in the world.
Computer Science
Computer Vision
Minimally Recognizable Configurations (MIRCs)
Published:
“Minimal Recognizable Configurations Elicit Category-selective Responses in Higher Order Visual Cortex” introduces the concept of Minimal Recognizable Configurations (MIRCs), as presented by Shimon Ullman et al. A MIRC is defined as the smallest visual configuration of image components that remains reliably recognizable as belonging to a specific category.
Connectome
Extracting the Most Predictive Subgraphs from the Human Connectome
Published:
This post is a summary of my Master’s thesis, Extracting Most Predictive Subgraphs From Models of Human Brain Connectivity, submitted in November 2020 for my M.Sc. in Life Science Informatics at the Bonn-Aachen International Center for Information Technology (B-IT), University of Bonn. Examiners: Prof. Dr. Thomas Schultz and Prof. Dr. Holger Fröhlich; advisor: Mohammad Khatami. The full PDF is linked at the bottom.
Generating Structural Connectomes from the Human Connectome Project
Published:
The Human Connectome Project (HCP) provides high-quality diffusion MRI data for over 1,000 healthy young adults, acquired at 3T with 1.25 mm isotropic resolution and 90 gradient directions at b=3000 s/mm^2.
DIPY
Crossing and Kissing Fibers in Fiber Tractography
Published:
One of the central challenges in diffusion MRI tractography is that a single imaging voxel is 2-3 mm on a side, yet contains thousands of axons that may run in completely different directions.
Advanced Interactive Visualization in DIPY with FURY
Published:
DIPY uses FURY (Fast Unified Renderer for You) as its 3D visualization engine. Beyond static screenshots, FURY supports interactive scenes with sliders, opacity controls, and video export.
Parallel Transport Tractography with DIPY
Published:
Parallel Transport Tractography (PTT) is a probabilistic tracking algorithm that frames fiber propagation as a curve evolving along a parallel transport frame. Compared to simpler deterministic methods, PTT better handles fiber crossings and produces smoother streamlines.
GPU-Accelerated Tractography with DIPY and cuslines
Published:
For whole-brain tractography the bottleneck is usually the inner tracking loop: at each step you evaluate a model, sample a direction, and move the streamline. DIPY supports offloading this to GPU via the cuslines library.
Introduction to Local Fiber Tracking with EuDX
Published:
Local tractography follows fibers step by step, estimating the local fiber direction at each point from an ODF model. This tutorial covers a full end-to-end local tracking pipeline using DIPY: ODF fitting, stopping criteria, seeding, and the EuDX algorithm.
Groupwise Bundle Registration with DIPY
Published:
When comparing white matter bundles across subjects, they first need to be coregistered to a common space. DIPY’s groupwise_slr function does this without requiring a pre-defined atlas.
Reslicing Diffusion MRI Datasets to Isotropic Voxels
Published:
Many diffusion MRI acquisitions use anisotropic voxels, for example 2 x 2 x 3 mm. This can cause issues downstream in tractography and registration algorithms that assume isotropic spatial resolution.
Getting Started with DIPY
Published:
DIPY is an open-source Python library for the analysis of diffusion MRI data. In this first tutorial I walk through the basics: loading data, building a gradient table, and visualizing axial slices.
Deep Belief Networks
Restricted Boltzmann Machines for Interpretable Neuroimaging
Published:
This post is a write-up of a seminar report I wrote during my M.Sc. in Life Science Informatics at the University of Bonn, for the “Visualization and Medical Image Analysis” seminar at b-it (WiSe 2019/20): Using Generative-Discriminative Learning in Neuroimaging for Interpretable Predictions. The full report is linked at the bottom; this is the part of it I still think about most, Restricted Boltzmann Machines, and why a fairly old, physics-flavored idea turned out to be a useful tool for making neuroimaging classifiers less of a black box.
Diffusion MRI
Extracting the Most Predictive Subgraphs from the Human Connectome
Published:
This post is a summary of my Master’s thesis, Extracting Most Predictive Subgraphs From Models of Human Brain Connectivity, submitted in November 2020 for my M.Sc. in Life Science Informatics at the Bonn-Aachen International Center for Information Technology (B-IT), University of Bonn. Examiners: Prof. Dr. Thomas Schultz and Prof. Dr. Holger Fröhlich; advisor: Mohammad Khatami. The full PDF is linked at the bottom.
Generating Structural Connectomes from the Human Connectome Project
Published:
The Human Connectome Project (HCP) provides high-quality diffusion MRI data for over 1,000 healthy young adults, acquired at 3T with 1.25 mm isotropic resolution and 90 gradient directions at b=3000 s/mm^2.
Crossing and Kissing Fibers in Fiber Tractography
Published:
One of the central challenges in diffusion MRI tractography is that a single imaging voxel is 2-3 mm on a side, yet contains thousands of axons that may run in completely different directions.
Advanced Interactive Visualization in DIPY with FURY
Published:
DIPY uses FURY (Fast Unified Renderer for You) as its 3D visualization engine. Beyond static screenshots, FURY supports interactive scenes with sliders, opacity controls, and video export.
Parallel Transport Tractography with DIPY
Published:
Parallel Transport Tractography (PTT) is a probabilistic tracking algorithm that frames fiber propagation as a curve evolving along a parallel transport frame. Compared to simpler deterministic methods, PTT better handles fiber crossings and produces smoother streamlines.
GPU-Accelerated Tractography with DIPY and cuslines
Published:
For whole-brain tractography the bottleneck is usually the inner tracking loop: at each step you evaluate a model, sample a direction, and move the streamline. DIPY supports offloading this to GPU via the cuslines library.
Introduction to Local Fiber Tracking with EuDX
Published:
Local tractography follows fibers step by step, estimating the local fiber direction at each point from an ODF model. This tutorial covers a full end-to-end local tracking pipeline using DIPY: ODF fitting, stopping criteria, seeding, and the EuDX algorithm.
Groupwise Bundle Registration with DIPY
Published:
When comparing white matter bundles across subjects, they first need to be coregistered to a common space. DIPY’s groupwise_slr function does this without requiring a pre-defined atlas.
Reslicing Diffusion MRI Datasets to Isotropic Voxels
Published:
Many diffusion MRI acquisitions use anisotropic voxels, for example 2 x 2 x 3 mm. This can cause issues downstream in tractography and registration algorithms that assume isotropic spatial resolution.
Getting Started with DIPY
Published:
DIPY is an open-source Python library for the analysis of diffusion MRI data. In this first tutorial I walk through the basics: loading data, building a gradient table, and visualizing axial slices.
EEG
Combined EEG-fNIRS Pipeline: Multimodal Neuroimaging from Recording to Analysis
Published:
EEG measures the scalp potential generated by synchronised postsynaptic currents in cortical pyramidal neurons. It has millisecond temporal resolution but poor spatial resolution, and the signal is smeared by the skull and scalp.
Face Perception
Prosopagnosia: When Faces Refuse to Stick
Published:
There is a particular kind of disorientation in not recognizing a face you’ve seen a hundred times: a colleague, a neighbor, sometimes even your own reflection in a photograph. This is prosopagnosia, or face blindness, the inability to recognize faces despite otherwise normal vision and intact intelligence. The name comes from the Greek prosopon (face) and agnosia (not knowing), and it was first described by the neurologist Joachim Bodamer in 1947.
GPU
GPU-Accelerated Tractography with DIPY and cuslines
Published:
For whole-brain tractography the bottleneck is usually the inner tracking loop: at each step you evaluate a model, sample a direction, and move the streamline. DIPY supports offloading this to GPU via the cuslines library.
Graph Theory
Extracting the Most Predictive Subgraphs from the Human Connectome
Published:
This post is a summary of my Master’s thesis, Extracting Most Predictive Subgraphs From Models of Human Brain Connectivity, submitted in November 2020 for my M.Sc. in Life Science Informatics at the Bonn-Aachen International Center for Information Technology (B-IT), University of Bonn. Examiners: Prof. Dr. Thomas Schultz and Prof. Dr. Holger Fröhlich; advisor: Mohammad Khatami. The full PDF is linked at the bottom.
HCP
Generating Structural Connectomes from the Human Connectome Project
Published:
The Human Connectome Project (HCP) provides high-quality diffusion MRI data for over 1,000 healthy young adults, acquired at 3T with 1.25 mm isotropic resolution and 90 gradient directions at b=3000 s/mm^2.
MEG
MEG Beamforming: Forward and Inverse Problems
Published:
Magnetoencephalography (MEG) measures the tiny magnetic fields produced by synchronised postsynaptic currents in cortical neurons. Compared to EEG, MEG is less distorted by the skull and scalp, giving it better spatial resolution.
MNE-Python
Combined EEG-fNIRS Pipeline: Multimodal Neuroimaging from Recording to Analysis
Published:
EEG measures the scalp potential generated by synchronised postsynaptic currents in cortical pyramidal neurons. It has millisecond temporal resolution but poor spatial resolution, and the signal is smeared by the skull and scalp.
MRtrix
Generating Structural Connectomes from the Human Connectome Project
Published:
The Human Connectome Project (HCP) provides high-quality diffusion MRI data for over 1,000 healthy young adults, acquired at 3T with 1.25 mm isotropic resolution and 90 gradient directions at b=3000 s/mm^2.
Machine Learning
Restricted Boltzmann Machines for Interpretable Neuroimaging
Published:
This post is a write-up of a seminar report I wrote during my M.Sc. in Life Science Informatics at the University of Bonn, for the “Visualization and Medical Image Analysis” seminar at b-it (WiSe 2019/20): Using Generative-Discriminative Learning in Neuroimaging for Interpretable Predictions. The full report is linked at the bottom; this is the part of it I still think about most, Restricted Boltzmann Machines, and why a fairly old, physics-flavored idea turned out to be a useful tool for making neuroimaging classifiers less of a black box.
Extracting the Most Predictive Subgraphs from the Human Connectome
Published:
This post is a summary of my Master’s thesis, Extracting Most Predictive Subgraphs From Models of Human Brain Connectivity, submitted in November 2020 for my M.Sc. in Life Science Informatics at the Bonn-Aachen International Center for Information Technology (B-IT), University of Bonn. Examiners: Prof. Dr. Thomas Schultz and Prof. Dr. Holger Fröhlich; advisor: Mohammad Khatami. The full PDF is linked at the bottom.
Multimodal
Combined EEG-fNIRS Pipeline: Multimodal Neuroimaging from Recording to Analysis
Published:
EEG measures the scalp potential generated by synchronised postsynaptic currents in cortical pyramidal neurons. It has millisecond temporal resolution but poor spatial resolution, and the signal is smeared by the skull and scalp.
Neuroimaging
Restricted Boltzmann Machines for Interpretable Neuroimaging
Published:
This post is a write-up of a seminar report I wrote during my M.Sc. in Life Science Informatics at the University of Bonn, for the “Visualization and Medical Image Analysis” seminar at b-it (WiSe 2019/20): Using Generative-Discriminative Learning in Neuroimaging for Interpretable Predictions. The full report is linked at the bottom; this is the part of it I still think about most, Restricted Boltzmann Machines, and why a fairly old, physics-flavored idea turned out to be a useful tool for making neuroimaging classifiers less of a black box.
Extracting the Most Predictive Subgraphs from the Human Connectome
Published:
This post is a summary of my Master’s thesis, Extracting Most Predictive Subgraphs From Models of Human Brain Connectivity, submitted in November 2020 for my M.Sc. in Life Science Informatics at the Bonn-Aachen International Center for Information Technology (B-IT), University of Bonn. Examiners: Prof. Dr. Thomas Schultz and Prof. Dr. Holger Fröhlich; advisor: Mohammad Khatami. The full PDF is linked at the bottom.
Combined EEG-fNIRS Pipeline: Multimodal Neuroimaging from Recording to Analysis
Published:
EEG measures the scalp potential generated by synchronised postsynaptic currents in cortical pyramidal neurons. It has millisecond temporal resolution but poor spatial resolution, and the signal is smeared by the skull and scalp.
MEG Beamforming: Forward and Inverse Problems
Published:
Magnetoencephalography (MEG) measures the tiny magnetic fields produced by synchronised postsynaptic currents in cortical neurons. Compared to EEG, MEG is less distorted by the skull and scalp, giving it better spatial resolution.
Generating Structural Connectomes from the Human Connectome Project
Published:
The Human Connectome Project (HCP) provides high-quality diffusion MRI data for over 1,000 healthy young adults, acquired at 3T with 1.25 mm isotropic resolution and 90 gradient directions at b=3000 s/mm^2.
Crossing and Kissing Fibers in Fiber Tractography
Published:
One of the central challenges in diffusion MRI tractography is that a single imaging voxel is 2-3 mm on a side, yet contains thousands of axons that may run in completely different directions.
Advanced Interactive Visualization in DIPY with FURY
Published:
DIPY uses FURY (Fast Unified Renderer for You) as its 3D visualization engine. Beyond static screenshots, FURY supports interactive scenes with sliders, opacity controls, and video export.
Parallel Transport Tractography with DIPY
Published:
Parallel Transport Tractography (PTT) is a probabilistic tracking algorithm that frames fiber propagation as a curve evolving along a parallel transport frame. Compared to simpler deterministic methods, PTT better handles fiber crossings and produces smoother streamlines.
GPU-Accelerated Tractography with DIPY and cuslines
Published:
For whole-brain tractography the bottleneck is usually the inner tracking loop: at each step you evaluate a model, sample a direction, and move the streamline. DIPY supports offloading this to GPU via the cuslines library.
Introduction to Local Fiber Tracking with EuDX
Published:
Local tractography follows fibers step by step, estimating the local fiber direction at each point from an ODF model. This tutorial covers a full end-to-end local tracking pipeline using DIPY: ODF fitting, stopping criteria, seeding, and the EuDX algorithm.
Groupwise Bundle Registration with DIPY
Published:
When comparing white matter bundles across subjects, they first need to be coregistered to a common space. DIPY’s groupwise_slr function does this without requiring a pre-defined atlas.
Reslicing Diffusion MRI Datasets to Isotropic Voxels
Published:
Many diffusion MRI acquisitions use anisotropic voxels, for example 2 x 2 x 3 mm. This can cause issues downstream in tractography and registration algorithms that assume isotropic spatial resolution.
Getting Started with DIPY
Published:
DIPY is an open-source Python library for the analysis of diffusion MRI data. In this first tutorial I walk through the basics: loading data, building a gradient table, and visualizing axial slices.
Neuroscience
Prosopagnosia: When Faces Refuse to Stick
Published:
There is a particular kind of disorientation in not recognizing a face you’ve seen a hundred times: a colleague, a neighbor, sometimes even your own reflection in a photograph. This is prosopagnosia, or face blindness, the inability to recognize faces despite otherwise normal vision and intact intelligence. The name comes from the Greek prosopon (face) and agnosia (not knowing), and it was first described by the neurologist Joachim Bodamer in 1947.
Modeling Why the Necker Cube Won’t Sit Still
Published:
Most write-ups of visual illusions stop at showing you the picture: here’s a cube that flips, isn’t the brain weird. That’s true but incomplete, it doesn’t say why perception should flip at all, rather than just settling on one interpretation and staying there, or why the flips arrive when they do rather than on some fixed schedule. This post builds a small computational model of exactly that switching process, for the Necker cube specifically, and asks whether a handful of very simple neural dynamics can reproduce the statistical signature of real perceptual switching.
Minimally Recognizable Configurations (MIRCs)
Published:
“Minimal Recognizable Configurations Elicit Category-selective Responses in Higher Order Visual Cortex” introduces the concept of Minimal Recognizable Configurations (MIRCs), as presented by Shimon Ullman et al. A MIRC is defined as the smallest visual configuration of image components that remains reliably recognizable as belonging to a specific category.
Audio and Text Alignment
Published:
In Neuroimaging language experiments it is often required that an orthographic transcript of the text is well aligned with the audio.
Music and Neurogenesis
Published:
So what is Neurogenesis? Neurogenesis is the process by which neurones are generated from neural stem cells and progenitor cells. The generation of neuronal cells in the hippocampus is responsible for memory as well as emotions. Every day our brain is generating new cells (about 700 on an average) which encode our memories and the emotions follow with what we associate that particular memory with.
Perception
Modeling Why the Necker Cube Won’t Sit Still
Published:
Most write-ups of visual illusions stop at showing you the picture: here’s a cube that flips, isn’t the brain weird. That’s true but incomplete, it doesn’t say why perception should flip at all, rather than just settling on one interpretation and staying there, or why the flips arrive when they do rather than on some fixed schedule. This post builds a small computational model of exactly that switching process, for the Necker cube specifically, and asks whether a handful of very simple neural dynamics can reproduce the statistical signature of real perceptual switching.
Physics
The beauty of Physics
Published:
All science enthusiasts are like wanderers in a field who try to unravel the secrets of nature. Human curiosity is inbuilt and we all have innate desire to discover the world around us and be informed! Physics is the fundamental science branching into narrow pathways leading scientific discovery of the unimaginable, the unexpected, the exciting and electrifying.
Python
Blender Python Scripting for Visual Effects
Published:
Blender ships with a full Python API (bpy) that exposes nearly every feature of the interface: object creation, material nodes, particle systems, geometry nodes, and rendering.
Restricted Boltzmann Machines
Restricted Boltzmann Machines for Interpretable Neuroimaging
Published:
This post is a write-up of a seminar report I wrote during my M.Sc. in Life Science Informatics at the University of Bonn, for the “Visualization and Medical Image Analysis” seminar at b-it (WiSe 2019/20): Using Generative-Discriminative Learning in Neuroimaging for Interpretable Predictions. The full report is linked at the bottom; this is the part of it I still think about most, Restricted Boltzmann Machines, and why a fairly old, physics-flavored idea turned out to be a useful tool for making neuroimaging classifiers less of a black box.
Source Localisation
MEG Beamforming: Forward and Inverse Problems
Published:
Magnetoencephalography (MEG) measures the tiny magnetic fields produced by synchronised postsynaptic currents in cortical neurons. Compared to EEG, MEG is less distorted by the skull and scalp, giving it better spatial resolution.
Tractography
Crossing and Kissing Fibers in Fiber Tractography
Published:
One of the central challenges in diffusion MRI tractography is that a single imaging voxel is 2-3 mm on a side, yet contains thousands of axons that may run in completely different directions.
Parallel Transport Tractography with DIPY
Published:
Parallel Transport Tractography (PTT) is a probabilistic tracking algorithm that frames fiber propagation as a curve evolving along a parallel transport frame. Compared to simpler deterministic methods, PTT better handles fiber crossings and produces smoother streamlines.
GPU-Accelerated Tractography with DIPY and cuslines
Published:
For whole-brain tractography the bottleneck is usually the inner tracking loop: at each step you evaluate a model, sample a direction, and move the streamline. DIPY supports offloading this to GPU via the cuslines library.
Introduction to Local Fiber Tracking with EuDX
Published:
Local tractography follows fibers step by step, estimating the local fiber direction at each point from an ODF model. This tutorial covers a full end-to-end local tracking pipeline using DIPY: ODF fitting, stopping criteria, seeding, and the EuDX algorithm.
Groupwise Bundle Registration with DIPY
Published:
When comparing white matter bundles across subjects, they first need to be coregistered to a common space. DIPY’s groupwise_slr function does this without requiring a pre-defined atlas.
Travelling
Perception of Mooney faces
Published:
PDF of the paper Perception of Mooney Faces: Extreme Generalization through Inverse Rendering.
Visit to M2L Summer School
Published:
So, I have travelled to multiple conferences as a part of my research career, but the M2L summer school 2025 (8-12 September 2025) was a bit special for me. I enjoyed every bit of the Summer School. It was super exciting to visit the Mediterranean countries and also see the enthusiasm of undergraduates. Before the summer school started, we got an invitation to the slack and also the whole schedule of the summer school which was nothing short of great!
SfN Neuroscience Chapter
Published:
So after ICLR 2025 in Singapore, I visited the SfN (Society for Neuroscience Chapter) in Singapore at the A* Star institute which was nothing beyond amazing. The facilities at the institute were world class and I got to meet some of the best researchers in Neuroscience there, and alongside present my work! I was happy to have 20 people at my poster with great discusssions about something new that we were putting forward. I am glad I got the chance to visit Singapore for this conference as it was once in a lifetime opportunity for me!
First Conference at BCCN Berlin
Published:
As an undergrad, it was the first time I got a chance to attend a conference. It was the Bernstein Computational Neuroscience Conference and the International Conference on Computational Neuroscience which was held from 11–12 October 2017 at Humboldt Universität zu Berlin (Humboldt University of Berlin). It was an enriching experience for a novice like me to get acquainted with ideas of the top researchers in the field. Almost like a dream come true for me.
Vision
Prosopagnosia: When Faces Refuse to Stick
Published:
There is a particular kind of disorientation in not recognizing a face you’ve seen a hundred times: a colleague, a neighbor, sometimes even your own reflection in a photograph. This is prosopagnosia, or face blindness, the inability to recognize faces despite otherwise normal vision and intact intelligence. The name comes from the Greek prosopon (face) and agnosia (not knowing), and it was first described by the neurologist Joachim Bodamer in 1947.
Modeling Why the Necker Cube Won’t Sit Still
Published:
Most write-ups of visual illusions stop at showing you the picture: here’s a cube that flips, isn’t the brain weird. That’s true but incomplete, it doesn’t say why perception should flip at all, rather than just settling on one interpretation and staying there, or why the flips arrive when they do rather than on some fixed schedule. This post builds a small computational model of exactly that switching process, for the Necker cube specifically, and asks whether a handful of very simple neural dynamics can reproduce the statistical signature of real perceptual switching.
Visual Effects
Blender Python Scripting for Visual Effects
Published:
Blender ships with a full Python API (bpy) that exposes nearly every feature of the interface: object creation, material nodes, particle systems, geometry nodes, and rendering.
Visualization
Advanced Interactive Visualization in DIPY with FURY
Published:
DIPY uses FURY (Fast Unified Renderer for You) as its 3D visualization engine. Beyond static screenshots, FURY supports interactive scenes with sliders, opacity controls, and video export.
fNIRS
Combined EEG-fNIRS Pipeline: Multimodal Neuroimaging from Recording to Analysis
Published:
EEG measures the scalp potential generated by synchronised postsynaptic currents in cortical pyramidal neurons. It has millisecond temporal resolution but poor spatial resolution, and the signal is smeared by the skull and scalp.